Sabermetrics applying to Soccer

[beineke]

Originally posted by maxim-1
Don't nessicarily knock that because the point wouldn't be to say, oh look how stupid this guy is using statistics to do something stupid. The point would be to say, "was he right?" Let's find some numbers to either back this guy up or find out if he's wrong. If he's right, then why's he right? What could be the reasons, and is that statistically important or just a nature of the game?

This is a very interesting point for conversation.

Actually, Egil Olsen of Norway used the "four-passes" insight to develop a strategy that turned his country from a perennial doormat into a team that qualified for the 94 WC, 96 European Championship, and 98 WC.

No time to discuss this in detail right now, though ... maybe over the weekend.

[the101er]

Well, as you can see, I have more questions than answers. But it's great to have a place to bring those questions and start looking for some concrete answers, rather than just "gut feel".

I'm learning. For example, the "gut feel" reaction to the theory of "direct play" was that it was debunked by the Brasilians winning two World Cups. But, has anyone studied how Brasil scored their goals in those games?

The British magazine "FourFourTwo" ran an article this month showing the effect on results if games were played up to just the 85th minute. Undoubtedly, there are flaws in the analysis, and I'm not smart enough to say what. But it shows there is an interest in debunking the myths of soccer like: "goals scored in the last 5 minutes are the most important".

I can see why this needs to be a whole forum, rather than one thread. There are just so many interesting areas to start looking at.

[beineke]

Originally posted by the101er

I'm learning. For example, the "gut feel" reaction to the theory of "direct play" was that it was debunked by the Brasilians winning two World Cups. But, has anyone studied how Brasil scored their goals in those games?


Remember that Brazil has gone to a much more physical style of play, which may have contributed to their strong results.

Let's get back to Olsen and the "four passes or less" study. First, note that this was observed in European international play. That's important because we're talking about teams that don't play together much. They aren't able to refine their combination play in the attack, so they need to be more opportunistic.

Now, what did Olsen see that was leading to goals?

1. Giveaways Opponents would pounce on a defensive misplay or misunderstanding. To avoid this problem, Olsen made sure that his own defenders communicated well and didn't screw around in the back.

2. Bad spacing International defenders tended to cover the short passes ok, but far-away defenders were easy to catch out of position (perhaps a symptom of unfamiliarity with their national team's system). Olsen made sure that his own defenders were very disciplined, and he encouraged long passes to probe for breakdowns on the other side.

3. Knockdowns It ain't pretty, but a high ball into the area can get deflected in a whole variety of ways. The defense can't cover all of them.

At one point, Olsen expressed bafflement at the fact that England manager Graham Taylor wasn't employing a direct style. Olsen's perspective proved correct -- in the Qualifiers for USA 94, Norway eliminated England by playing classic English soccer.

[microbrew]

Olsen's observations leads me to think:

Let's compare international statistics to club statistics. Intuitively and qualitatively, most observers of the game can tell that there's a difference.

Also, does Olsen's observartions imply that more goals are given up in international play, in that defenses are not as strong?

[superdave]

Originally posted by the101er
Soccer coaches do try to be as logical as possible. For example, Anson Dorrance claims to measure everything measurable. He keeps score in all scrimmages, sprints, fitness tests and tries to measure his players as objectively as possible.
I'll bet you a million dollars he got this from Dean Smith.

[kenntomasch]

From SoccerBasher Illu...I mean Sports Illustrated, 12/7/98 in a story on Dorrance (can you believe the Keller thing has been going on for five years now?):


"...Taking a cue from Dean Smith, he kept a detailed record of every player's performance in practice and posted the results. Intensity kicked up a notch, and younger players refused to genuflect to their elders."

Dave, do you take PayPal? ;)

[superdave]

Nah, but if you had any idea how influential Dean is/was on the UNC campus, and in the sports department, it wasn't hard to figure out.

His book, A Coach's Life, is a great read. Not alot of "we won this game, then lost the next game" stuff. Just thoughts about, well, a coach's life.

[mpruitt]

know what might be interesting as sort of a coaches question to answer. what do the types of drills that coaches run in practices say about the game. i know that this would be a really hard thing to qualify because there are a ton of variables. obviously they would include the coaches style, personality, opponent etc etc etc. however, wouldn't the best indication of the skills and attributes that are important in the game reflect on the most important aspects of it. maybe even taking a look at some youth development thing in tersm of individual players would help. reason being that if theres a general conscensus as to what's important then you could either find a way to test if it's valid or try to find a way to quantify it.

On a different note: A wuestion was asked by Huss in regards to where our new forum would possiable go. Please post if you have any suggestions or comments, another question to answer would be if you think it'd be best to have the forum appear on Bigsoccer Live, ie. the main board. Question would be having it viewed more or having it be more of an exclusive thing for those who would be more truly interesed?

[kenntomasch]

More on the 2-0 lead, assuming I didn't make any mistakes doing it this late:

In 2002, there were 52 games in which a team took a 2-0 lead. They won 50, lost 2 and tied 0. That's a .962 W-L% and there were only 3 blown leads (those two losses and one other in which the team in question won anyway).

So from 1999-2002, I have 271 games in which a team took a 2-0 lead. They won 248 games outright, or 91.5%.

I show only 20 blown leads in those 271 games, or 7%.

So your chances of actually losing or blowing the lead don't seem to be enough to call a 2-0 lead "the most dangerous lead" (at least in MLS, it would be interesting to see if that holds up, say in the A-League, WUSA, and college soccer), but I'll bet you'll still hear it said whenever a team does blow one.

[beineke]

Originally posted by microbrew

Also, does Olsen's observartions imply that more goals are given up in international play, in that defenses are not as strong?

Couple of things...
1) That study was ten or fifteen years ago. In the meantime, a lot of the UEFA minnows have learned how to tighten up their defenses. That's why you rarely see blowouts in European competition any more.

2) Even though defenses are not as strong, I think that offenses suffer even more in international play. Part of the reason that teams didn't score off of possessions with 5+ passes is that the attacking players aren't familiar enough with each other to develop good scoring chances.

Incidentally, I think it's quite common to see a learning effect from these well-publicized studies. As I recall, the 98 World Cup had an awful lot of goals being allowed from re-starts. Then in WC02, there were very few goals from re-starts (the USA being a notable exception). Clearly, most teams had emphasized their defense against opposing set pieces.

[superdave]

Originally posted by maxim-1
On a different note: A wuestion was asked by Huss in regards to where our new forum would possiable go.
To me, it should go into the "on the field" forum, alongside the referee, coach, and player forums.

[boffo]

Wow, this is an amazing thread (geek that I am), and I hope somewhere down the line that the subject does get its own forum. As the posts here show, there is quite a bit to say on the subject. I have some ideas of my own, but here are just a few culled from the month of posts here.

1. Getting back to one of Voros' original points (and dredging up an earlier debate), it is not so much that subjective data are to be rejected in favor of objective data, but that they need to be handled somewhat differently. In the social sciences this is an obvious problem, where many of the things being studied are subjective. One solution is to use multiple people to measure the same event independently of one another and to calculate the degree to which they agree. There is quite a literature on "intercoder reliability" which deals with this question. The problem for us soccermetrics fans on zero budgets is that this is a labor-intensive approach. (This seems to be what Maxim-1 is talking about doing with Matchtracker.)

1A. Another is to come to the coding of subjective data with a well-developed theory justifying your coding procedure, perhaps based on one's understanding of soccer play. That way, if someone questions your coding they can turn to your theory, point out where they think you went wrong, and offer another theory and coding scheme. Such is the nature of real research, and that's what I see starting to happen in this thread. We have to resign ourselves to the understanding that we will in all likelihood not find a definitive measure or set of measures for what we care about proving our point with certainty (Popper says we won't), but as Voros pointed out at the beginning that is no reason not to try.

2. A problem a number of posters have danced around is the issue of selection bias. This occurs when you seek to explain a certain kind of outcome -- e.g., when do goals get scored -- and ignore other outcomes. Voros' point earlier about the number of passes leading to goals vs number of passes in general is right on point: If the number of consecutive passes is HIGHER in the run of play than it is when a goal is produced, then this is powerful counter evidence. We have to remain agnostic about the question and not gather data in a way that loads the answer we want, however inadvertently. (To give a standard non-soccer example: If we want to explain why certain people default on loans, we need to take into account who gets a loan and who doesn't in the first place, because presumably that earlier step is meant to weed out likely defaulters. Otherwise, asking only about who defaults produces hopelessly misleading results.)

3. Multiple regression will have to be the answer to many of the questions we are asking. Simple tabulations or correlations between two variables can be useful, but as several people have pointed out along the way there may be alternate causes that we need to control for. By considering these alternate causes together can we sort out the competing theories. Also, it is valuable to know whether multiple theories are correct -- entirely possible -- and whether one is stronger than the other.

4. I want to urge caution on all of you who try to do this in Excel, however. There have been quite a few studies in the statistics and econometrics literatures, and Excel tends to be less reliable than actual statistical software (SPSS, Stata, SAS, etc), sometimes dramatically so. Put simply, sometimes the Excel results are just plain wrong. If you are doing anything slightly sophisticated like regression, you are much better off using statistical software, if available.

[Karl K]

I am very very late to this thread...but having read all the posts, and reviewed some of the terrific links to papers and other articles, I am glad I did.

And I hope we can get a separate forum for this subject.

OK, now, in the spirit of full disclosure, I will confess that this is going to be a monster post. So be forewarned.

First I want to talk about our efforts at player performance charting. Second, I want to provide some important broad characterizations of "soccermetrics" that I think will be useful to segregate future discussions, assuming we do get our own forum. And, finally, I want to put forth a few of my (admittedly untested) theories about what IS important in the game, and what can and should be measured.

Topic 1. Player Performance Charting

As Beineke pointed out, we tried to do what I called "Player Performance Charting" for J/K 2002. The web site I put together to publish the information, along with some seminal work done by Ric Miller, is still up.

See:

http://www.playerchart.freewebsites.com/

The process of doing player performance is very labor intensive, which is why it was so hard to get follow through. Which was a shame, because I had built an entire methodology, and a Microsoft Access database to gather up all the information, and to create some interesting reports.

This effort was originally inspired by the work of Ric Miller, whose now defunct site, “The Game Behind the Game” was really fascinating and stimulating. Basically, Ric argued that many post-game assessments of player performance were highly impressionistic in nature, and therefore fraught with all the dangers of subjectivity. In an effort to put some precision behind player observations, Ric put together a very interesting shorthand. All of Ric’s shorthand tracked just “ball touch” events. I tried to take his foundation and go beyond in much more detail.

Basically, Player Performance Charting is designed to create a record of player actions during a game. This record, or chart, is created from detailed videotape watching and recording events. These recorded events are then run through a Microsoft Access database to produce the reports.

The charting system I devised is designed to track mostly "on the ball" events in the game -- number of touches taken, passing, dribbling, shooting, crossing, tackling, etc. In addition, we made provisions for certain "off the ball" events such as runs made, shields to retain possession, and situations where defensive pressure and cover is applied.

Besides quantifying such actions, game charters had the option of "qualifying" such events. Was this a good dribbling sequence? Or bad? Was the pass lost an especially bad giveaway or acceptable give the dynamic of the game, such as an ambitious forward pass to a front runner?

From our charting efforts, we wanted to produce three reports. The first, the Game Narrative, is a minute by minute listing of what the players did, in the order in which they did them. It chronicles the type of events in the "event sequence" (a "one touch, then "a completed pass" would be two events in an event sequence). It tells us what area of the field this event sequence occurs (defensive, middle or attacking third; or defending or attacking area). And there is a space for optional commentary on the event sequence.

The second report is the Narrative by Player. This report shows us the event sequences of each player separately; you can look at this report and see, for example, all of DaMarcus Beasley's events over the course of the game.

The third report is Player Summaries, and really the most interesting report. In the summaries we total up for each player (and the team as a whole), key event types -- such as how many touches on the ball did this player get? How many tackles won and lost? How many corners created...or corners conceded? And the like.

I’ll talk about the shortcomings of this system in a minute, but let me say this. The information this system provides can be VERY revealing about an individual player’s performance particularly. Remember back in 2001 when many were hot to have Jeff Cunningham get caps? Well, I charted his performance in a game against the Dallas Burn during the summer of 2001, and I was just shocked at the limitations in Jeff’s game. He won very few balls, checked to ball rarely, played little defense, made few combinations passes. Like the most naïve youth soccer player, all he wanted to do was run off, get the ball at his feet, face up to goal, and dribble.

As in the case of Cunningham, the charting system we created can tell us some key things about a player or a team's performance. For example, it can give us a good indication of how "active" a player is -- is he getting touches on the ball, is he winning tackles, is he completing his passes to teammates? If he is a defender or defensive midfielder, is he winning balls, making good clearances? If he is an attacking player, is he getting good touches in the attacking third or in the attacking area? Are his dribbling sequences good?

The system can answer certain kinds of questions – some of which were the source of endless (some might say monotonous) debate about the qualities of certain players. . For example, does Claudio Reyna spend most of his time in the middle or defensive third retrieving balls...or does he "get into the attack?" Our charts were designed to answer this question.

But because soccer is a game where so much happens off the ball, there are things that the charting system can't tell us. For example, it won't show whether a player has taken up a good position off the ball, or whether the team's defensive "shape" is strong, of if the team created good "depth" on the attack. And a "pass completed" may in fact be a "hospital ball" from Player A that puts teammate Player B under duress, forcing Player B to whack the ball desperately to get out of trouble (though there the game charter has the option of qualifying the nature of this pass).

Over the course of a game, though, these ambiguous situations should be set in the context of a longitudinal performance -- most players are going to get anywhere from 35 to 70 touches per game. Meanwhile, to deal with these situations, and to add richness to the charting, game charters had the option of adding brief comments to qualify events and explain what happened. That's why we always wanted the Player Summary reports to be be viewed in the context of event comments.

I would love to resurrect this effort for the USMNT, particularly in qualifying and the runup to the next world cup. Take a look at what we’ve done and send me a PM if you want to get this up and running again.

Topic 2. Soccermetrics Categories

I think it would be useful, going forward with the notion of soccermetrics, to view this subject matter according to three key categories. There is some overlap here, but I th ink these distinctions are useful. If anyone can think of more useful or revealing labels, chime in of course.

Player/Team Performance – These are soccermetrics related to individual player performance in games, with individual performance items rolled up into team totals. A more advanced boxscore, as it were. The Player Performance Charting we did would fall into this category, as would Matchanalysis.com and OPTA.

Multi-Game Metrics – These would be studies/statistics on issues and ideas that cut across teams, games, seasons. For example, whether or not, say, the timing of goals scored in World Cup tournament follows a Possion distribution, or whether teams that have more shots, regardless of whether they are on frame, score more or win more.

Management Metrics – These would be studies/statistics on issues related to the decisions that managers make, issues such a player personnel selection, use of the draft/college players/P40 players, age of team roster, success of teams with a preponderance of national team players, and so on.

Topic 3. Some (untested) views on what is really important to success in the game today

Here they are, for what it’s worth. If we could figure out ways to measure and analyze these issues with some rigorous mathematical methodology….well, that would be something.

--Athleticism is beginning to trump skill – I ‘d love to see the 40 times of our national team players, vs. those of, say, England or France. Or Vox2 measurements. Speed, flexibility, fitness, these things are key in the modern game.

--Winning teams have great defensive organization. Rarely out of shape, always strong pressure cover, giving teams few chances at goal, win balls all over the field, allow few shots.

--Good teams win head balls in the penalty area on both ends of the field.

--Good teams have quality touches inside the penalty area, and prevent other teams from having quality touches in the penalty area.

--Good teams “hold” the ball – by that I don’t mean dominating possession. It’s that when you HAVE it, you do something productive WITH it.

--Good teams cause turnovers, but don’t commit them. Especially in the middle and final third.

--Teams that switch the point of attack and use the entire field when in possession are better than teams that are “sided” in their ball possession.

--Teams that can complete a higher percentage of on-the-ground passes than the other team (not quantity, but percentage) will do better on average.

--Teams that on average complete longer on-the-ground passes than the opposition do better.

--Good teams convert a high percentage of their numbers-up scoring opportunities. Weak teams have fewer, and don’t convert them at the percentage that better teams do.

--Good teams have coaches who exploit advantageous player matchup opportunities, and are adept at hiding playing matchup weaknesses.

***************

Anyway, some ideas to chew on. Let’s hope we can get our own forum.

[mpruitt]

Goodness. That was a monster post in more than a few ways. I'm not even going to respond to it till i take some real time to chew threw it. At the start of this thread someone had mentioned that a bigsoccer poster had tried to organize something like this before, but i'm suprised i didn't come across your site. Lord knows that i searched.

Please mke sure to post in the Suggestions forum in regards to your support and general ideas as to what it should represent. Stats and Analysis Forum request thread (http://www.bigsoccer.com/forum/showthread.php?s=&threadid=64954&perpage=15&pagenumber=2)

My initial reaction to the things you talk about is that maybe in an effort to take the subjective nature out of the game you may be putting more into it. obviously the lengths in which you've gone to try to break the game down are amazing but i wonder if it's not too specific. Either way absolutely great effort and i'm going to take some serious time to read that over.

On a different note:
What amazes me about this thread is that we've obviously found that there are enough people interested in this type of thing, but there hasn't been any kind of organizing factor behind it to bring all of it together at least in some small way. Obviously you have people like karl, who thought these things were compelling to put some work into it. Then there's people like Voros who could easily represent a sort of crossover between baseball and soccer. As well as, folks at Matchtracker and RSSF.org who already have a bit of a foundation in place. Finally, people like most of us who are obsessive about the game and curious if there isn't enough being done to understand it. I really think that hopefully a forum surrounding this topic is a really important thing as it will maybe be the first step in creating some sort of Soccer Researchers of America organization.

[microbrew]

"Statistics and Analysis" in "Business, Media and Entertainment".

http://www.bigsoccer.com/forum/forumdisplay.php?s=&forumid=806

[SJFC4ever]

Originally posted by kenntomasch
It's just not as easy to quantify those things as it is in baseball or football.

I mean, it can be done (OPTA does a bunch of it) but it's labor-intensive.

http://msn.skysports.com/skysports/article/0,,0-1027137,00.html


I think Goalkeeper is a position where stats could be used fairly well. It is a position that is largely independent of the others, since the keeper is the only player permitted to handle the ball (inside the penalty area).

Save % is an obvious stat, but I think you would have to adjust for something like save difficulty, which would be based on range, angle, velocity and height.

Oops, maybe I should post this in the new forum.

NB I have reposted this, in "goalkeeper performance?" thread.

[monster]

I'm moving this over to the new forum, but willleave a redirect in B&M for those looking for it.

Congrats on the new home.

[Nutmeg]

Woohoo! New home.

[Andy_B]

I am obviously late to the party here.

I did my best to briefly read the thread but I am going to have to go back and digest it deeper later.

I have two quick points that are related.

Unless I missed it, I am surpised no one besides Karl mentioned Opta, the granddaddy if you will of soccer stats.

On a related note, there was talk early in the thread about how it would be hard to get people to believe in stats.

If I am not mistaken, this transformation has already been taking place in England for a few years now thanks to Opta. I have read articles in which managers refer to the Opta system in which they review with each player after every game to identify strengths and weaknesses for the individual player as well as the team.

I have purchased more than a few Opta yearbooks and they are nothing short of amazing considering how little there is for soccer stats out there. Think of a book the size of the Rothman's Yearbook, only packed with colorized stats and graphs for every team and every player and pretty much every touch of the ball for the entire season for the entire Premier League.

Andy

[Karl K]

Andy, thanks for the reference to the OPTA yearbook.

Do you have a link handy where you can purchase it?

I have found some shortcomings in what I have seen from OPTA, though it looks like I haven't seen everything.

One of the ways you could view our efforts at Player Performance Charting was to supply OPTA-like stats for the USMNT. Matchanalysis.com was doing something like that, but they have gone off in a different direction now, so, really, there's nothing out there for us American fans.